Subject: Re: [vserver] Small flaw inside guest on kernel 4.1.13-vs2.3.8.3
From: Urban Loesch <bind@enas.net>
Date: Fri, 19 Aug 2016 16:17:28 +0200

Hi Herbert,

thanks for your fast response.

Am 19.08.2016 um 15:06 schrieb Herbert Poetzl:
> On Fri, Aug 19, 2016 at 12:26:15PM +0200, Urban Loesch wrote:
>> Hi Herbert,
>
> Hey Urban!
>
>> I tried kernel 4.1.27 with patch 2.3.8.5.2.
>> Now the "free" command inside the guest shows the
>> assigned amount of memory to the guest.
>
>> But know it show always the same amount, also on
>> line "-/+ buffers/cache":
>
> You mean the buffers/cache information follows the
> memory information regardless of the actual usage?

Yes. Sorry if I explained it not exactly.

>
>> ...
>> linux-image-4.1.27-vs2.3.8.5.2-em64t-efigpt_1_amd64
>> ...
>> root@dbserver:/ # free
>>              total       used       free     shared    buffers     cached
>> Mem:      50331648   29124200   21207448          0          0          0
>> -/+ buffers/cache:   29124200   21207448
>> Swap:      4194304     171660    4022644
>> ...
>
>> Now it seems that the guest uses it's hole assigned
>> memory and this triggers my monitoring.
>
> Hmm, I don't see that, there is rougly the same
> amount of free memory as used memory in your example,
> so it looks like half the assigned memory is used.
>
> Please elaborate ...

Yes, this is because the guest was running for some days. When the amount of used memory
becomes to high I get an alarm. Then I increased it from 32GB to 50GB. Therefore the
example above
shows the usage after I increased the max memory cgroup.

It is not normal that the buffers/cache information shows exactly the
same as the memory information.

>
> Best,
> Herbert

Thanks
Urban

>
>> Thanks
>> Urban
>
>> Am 02.02.2016 um 10:34 schrieb Herbert Poetzl:
>>> On Mon, Feb 01, 2016 at 03:02:28PM +0100, Urban Loesch wrote:
>>>> Hi,
>
>>> Hey Urban!
>
>>>> there is a small flaw inside the guest on kernel 4.1.13-vs2.3.8.3.
>
>>>> The "free" command shows the full memory of a host and not the
>>>> assigned amount, even if the "flags" file in
>>>> "/etc/vservers/$VSNAME/flags" is set like this:
>>>> ...
>>>> VIRT_MEM
>>>> VIRT_CPU
>>>> VIRT_LOAD
>>>> ...
>
>>>> Some other details:
>>>> kernel:  4.1.13-vs2.3.8.3
>>>> util-vserver:
>>>> ii  libvserver0        0.30.216-pre3120-jessie0.1-1 amd64 dynamic
>>>> libraries for util-vserver
>>>> ii  util-vserver       0.30.216-pre3120-jessie0.1-1 amd64 utilities for
>>>> managing Linux-VServer guests
>>>> ii  util-vserver-build 0.30.216-pre3120-jessie0.1-1 amd64 tools which can
>>>> be used to build vservers
>>>> ii  util-vserver-core  0.30.216-pre3120-jessie0.1-1 amd64 core utilities
>>>> of util-vserver
>>>> ii  util-vserver-sysv  0.30.216-pre3120-jessie0.1-1 amd64 initscripts for
>>>> util-vserver
>
>>>> Host-OS:	Debian Jessie
>>>> Guest-OS:	Debian etch oder higher. Also Jessie is affected.
>
>>>> The "top" command shows the correct assigned cpu eg. 14-15.
>>>> ...
>>>> top - 14:59:26 up 2 days, 20:25,  0 users,  load average: 0.05, 0.07, 0.02
>>>> Tasks:  16 total,   2 running,  14 sleeping,   0 stopped,   0 zombie
>>>> Cpu14 :  1.7%us,  0.0%sy,  0.0%ni, 98.3%id,  0.0%wa,  0.0%hi,  0.0%si,
>>>> 0.0%st
>>>> Cpu15 :  1.5%us,  0.0%sy,  0.0%ni, 98.5%id,  0.0%wa,  0.0%hi,  0.0%si,
>>>> 0.0%st
>>>> Mem:  32935924k total, 32745520k used,   190404k free,   248248k buffers
>>>> Swap:  7811068k total,     8620k used,  7802448k free,        0k cached
>
>>>>  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  P COMMAND
>>>> ...
>
>>> Thanks for reporting, but this is known, as the relevant
>>> code in the 4.1.x patches has not been adapted to the
>>> changes in the kernel (i.e. it is disabled for now)
>
>>>> All limits are assigned with cgroups.
>>>> Have you some idea how I can fix this?
>
>>> By adapting the relevant code (it is still there but
>>> disabled for now) to the changes in the kernel.
>
>>>> As I just said, this is not a real error, but only a flaw
>>>> that bothers me.
>
>>> Yeah, you are probably not the only one who misses this
>>> feature, I will see what I can do to get it working again.
>
>>> All the best,
>>> Herbert
>
>>>> Thanks and regards
>>>> Urban
>
>
>